New diagnostic tools leverage the power of Artificial Intelligence.
Too Much Data to Process
By 2030 all Baby Boomers will be over 65, which means that health-related issues become more and more top-of-mind for this aging population. Regular cancer screenings are one of these concerns. The sheer volume of such screenings already taxes existing healthcare systems. However, laboratories, diagnostic technicians, and healthcare providers are using powerful new technological tools to aid them in the work of helping patients live happier & healthier lives.
One such tool is Artificial Intelligence, commonly referred to as AI.
Unlike its counterparts depicted in the movies as sentient neural networks whose sole purpose is to destroy humanity, real AI has been a computing and data processing resource staple for decades. AI is as quotidian as the electric power grid and supermarkets.
Everything from predictive weather modeling to aid meteorologists to CAD-based generative design for engineers, AI has proven to be a powerful tool for many industries in an everyday capacity.
In the case of healthcare, data modeling and data processing have become synonymous with AI-driven environments capable of handling such massive volumes.
Take for example liquid biopsies to better predict infant cancers. The data associated with these tests are referred to as high-throughput data. Making connections is essential between high-throughput data on orders of magnitude within a smaller outcome sample space of patient responses. The results of these AI-driven computations expedite determinations on whether or not they have cancer.
Statistical models are useful for summarizing and describing variations to predictive models, and machine learning AI leverages these summaries that can make for more useful predictions, as seen above.
Imaging for Data Collection and AI Processing
From X-Rays, to CT (CAT) Scans, to MRIs, in vivo imaging technology has been one of the most powerful medical diagnostic tools in the modern era.
And, this is where AI provides a huge beneficial impact on human health. For example, a patient gets a standard set of images, AI-powered algorithms go through and look for abnormalities, such as tumors, breaks, enlarged hearts, or port placements to name a few.
AI then groups the images for humans to further examine them (considered the Gold Standard in diagnostic imaging analysis).
Imaging scan results, such as low-dose spiral computer tomography (LDCT), are grouped by AI accordingly:
- Radiographs that are of no concern
- Radiographs that might be of concern
- Radiographs that are definitely of concern
The hot topic today is radiology and radiographic analysis representing a wealth FDA approvals.
And though radiography provides a small transition point for AI to flourish in, it is by no means insignificant, nor siloed in that singular use case.
Furthermore, this specialty has become quite an attractive field—owing to the huge amount of high-dimensional data.
The biggest concern is that AI will replace human radiologist; however, that is an unfounded worry. The real concern is that radiologists who do not have AI-related skill sets will eventually be replaced by those who do.
AI vs Quantum Computing vs Machine Learning
One the most prevalent misconceptions in this field is that AI, Quantum Computing (QC), and Machine Learning (ML) are all interchangeable concepts. That is not the case. Essentially, AI is the entire universe of which both QC and ML are parts thereof.
The following simplified definitions might help clear this up:
AI – A field which combines computer science and robust datasets to enable humanlike problem-solving.
QC – A computer which makes use of the quantum states of subatomic particles to store information with blazingly fast computational speeds.
ML – Involves computers discovering how they can perform tasks without being explicitly programmed to do so.
For everyday activities such as high-throughput data processing, these technologies can be thought of as supervised, unsupervised, and semi-supervised.
Supervised is more common. It is a comfortable technology. For example, you get together your training set (i.e., these are the right answers). Did the AI get the right answer? If so, that’s called training, testing, and tuning.
Unsupervised is where you let the algorithms find patterns for itself. It is the least controlled. It is probably the scariest (think Skynet from the Terminator movies), but it is potentially the most useful. Why? Because computers see things differently than humans. And that change in perspective can lead to innovations and breakthroughs in all fields—especially medicine.
Other cutting-edge uses for AI-driven technologies are Deep Learning, AI Neural Networks, and Decision Trees (where decisions that need to be made, AI can be useful to facilitate those).
Roadblocks to Widely Adopting AI in Medicine
Probably the biggest challenge that AI faces with widespread adoption in the medical field is arguably unfamiliarity.
“AI is still considered a “black box” technology,” according to Dr. Christie Bergerson, a leading scientist in the field of QC who is a voting member of the ISO Committee on AI Standardization and Regulation, “not a lot of professionals are savvy to its uses and potential as a diagnostic tool for furthering human health. People don’t trust what they don’t understand (like the aforementioned unsupervised algorithms), which can be greatly exaggerated when it comes to healthcare. When a problem comes up, there’s not a decision-maker in the room that says, ‘Hey, AI could solve this problem.'”
Dr. Christie Bergerson – Leveraging Artificial Intelligence in Human Health (Webinar), University of Maryland, 2020
When a problem comes up, there’s not a decision-maker in the room that says, ‘Hey, AI could solve this problem.’”
Take validation of data models for example. A lot of education still needs to be done on how we validate these. Predictive models that aren’t validated—meaning people can’t replicate what the original authors of the study did—resulting in the data set was different from the model used in the study (i.e., unknown confounders causing something not to validate). This lack of reliability complicates the administrative case for next-level buy-in of adopting AI platforms in healthcare best practices. However, it is not AI’s fault but rather the mishandling of the data from the user’s perspective.
The Future of AI in Healthcare
Complex problems that basically regular computer methodology and systems cannot solve would greatly benefit from AI algorithms thrown at them. Add in QC, powered by subatomic particles known as qubits that are used to store info, immensely increasing the speed of processing high-throughput data that traditional computers are hard-pressed to handle. Finally, delivering and storing these results on deep learning (ML) networks served up in a cloud-based platform accessible to medical experts across the globe. And, what we are looking at here is a true paradigm shift of real-time knowledge bases from which to pool resources that will find cures to humanity’s oldest biological scourges.
The future for AI in Healthcare looks very bright.